positiveSYS.SOURCE: GitHub• 2026-11-07T15:16:01Z
HyperFlow H3 Flash: 13.7B Compressed Video and Audio Diffusion Model with INT8 Quantization
HyperFlow H3 Flash compresses a 31B video+audio diffusion model to 13.7B parameters with INT8 quantization, enabling 6-step generation of 5.2-15s videos at 24fps with stereo audio. It achieves real-time performance on 4× B-series GPUs and supports multiple runtime frameworks like ComfyUI and stable-diffusion.cpp.
*** END OF TRANSMISSION ***